Branch a Gemini chat: compare an alternative from an earlier reply

Branch a Gemini chat to explore an alternative from a specific earlier response in a new conversation. Reopen the original and the branch separately, then save a comparison card that distinguishes expected checks from observations.

Content checked: 19 October 2026.

What does a separate conversation branch help you do?

Branch a Gemini chat when you want to explore an alternative from a specific earlier response in a new conversation. After this walkthrough, you can identify that response, open a branch, try one alternative and separately reopen the original conversation. Your deliverable is a two-conversation comparison card, not a restored version or proof of immutable history.

Imagine a freelance facilitator who developed a fictional orientation outline. The original conversation later expanded the opening exercise. The facilitator now wants a quieter alternative starting from the earlier outline, without confusing that alternative with the original continuation. This is a teaching scenario, not an account action observed for this article. Use harmless material you are permitted to work with.

AI means artificial intelligence. The artificial intelligence guide offers broader starting tasks. This one is about a particular conversation boundary: which earlier response you selected, which later continuation belongs to the original, and which new request belongs to the alternative. Different titles alone do not establish that those objects have been correctly separated.

Choose a suitable existing conversation

Use gemini.google.com on a computer, signed into a permitted personal Google account. Google’s current chat-management instructions require signing in. For personal accounts, the described history features require Keep Activity to be on. It must already be enabled for this exercise. No activity-setting change is included merely to make the steps available.

Choose an ordinary non-sensitive conversation that contains an identifiable earlier Gemini response and a subsequent continuation. Those two points are the input to your comparison. A conversation containing only a single final response does not provide the same before-and-later distinction. If you cannot identify both points, record the missing input rather than pretending the comparison has been completed.

Do not use a temporary chat. Its access boundary is covered by the temporary-chat walkthrough, which is a different task. Here you need an ordinary original conversation you can revisit. Work and school administration are outside this narrow personal-account exercise; do not substitute these steps for your organisation’s rules.

No client document, private message or real participant information is needed. Record your account type rather than its full email address. Do not start from somebody else’s account or from a public conversation link sent to you. The intended objects are your own permitted original conversation and the new branch created from a response within it.

Mark the earlier response and the original later continuation

Open the original conversation and scroll to the response where you want the alternative to begin. Confirm that it is a Gemini response, not merely one of your own messages. Google’s desktop instructions put the branching command in the menu below a response. The conversation-level menu is not the same control as the menu attached to your chosen response.

On your card, write the original conversation title and a short harmless recognition cue for the selected response. In the facilitator example, this could be “initial three-part orientation outline.” Separately note the original later continuation, such as “expanded group introduction.” These are fictional examples. Use only content you actually see when completing your own observation.

The second cue matters because two similarly titled conversations can otherwise be difficult to distinguish after branching. You want to recognise the original’s later material independently from the alternative you will request. A concise cue is enough; copying the entire transcript adds work and may reproduce information the exercise does not need.

Do not replace response-level branching with copying the final answer into a blank chat and call the operations identical. This walkthrough is specifically about the native branch command at a selected response. If that response cannot be identified, leave the starting boundary unresolved. No deletion, restoration, sharing or history cleanup is needed to establish this input.

Open the branch and request one alternative

Below the chosen response, open Menu and select Branch in a new chat. Google’s current computer instructions describe this as creating a new conversation branching from that point. Inspect what opens in your own account. If the command is unavailable or you cannot distinguish the resulting conversation, record that stage as not checked instead of substituting a different operation.

In the new branch, ask for one bounded alternative. For the fictional outline, you could write: “From this initial outline, suggest a quieter opening based on individual reflection. Do not expand the remaining sections. This is a fictional planning exercise with no real participants or meeting details.” The answer is not supplied in advance; read the response you actually receive.

If the conversation names are confusing, use distinguishable labels such as “Practice 02 | Original outline” and “Practice 02 | Quiet opening.” Renaming is only an identification aid here. It is not the deliverable, and a different label does not by itself prove that you branched from the intended response. The selected starting point and alternative continuation must also be identifiable.

Branching is not a request to roll the original back or overwrite its later messages with an earlier version. You are exploring an alternative in a new conversation. Nor does a visibly separate conversation establish isolation of model memory or all account data. Do not turn an interface observation into a privacy, legal-compliance or retention guarantee.

Reopen and distinguish the two conversations

After reading the alternative response, separately open the original conversation from the side panel. Locate the later continuation you recorded before branching. Then separately open the branch and locate your alternative request and its response. Record the two reopen checks individually. Finding one conversation does not establish that you successfully inspected the other.

An expected check is not an observed result. No account, branch creation, generated alternative, two-chat reopening or timing trial was performed for this article. Fill the observation fields only from your own actions. If a connection problem prevents one reopening, leave that result incomplete. Do not infer a product-wide failure from an unexplained local obstacle.

Seeing the same initial outline in two places is not sufficient evidence of divergent continuation. Locate the alternative request itself and distinguish it from the original later exchange. If both windows seem to show the same conversation, do not resolve the uncertainty by changing titles alone. Return to the recognition cues and keep the status unresolved if identification still fails.

The public-link review concerns a different object: an existing shared snapshot. This exercise creates no public link and requires no recipient. Google’s Privacy Hub explains separate data and activity controls; it does not make a conversation branch proof of privacy or memory isolation. Keep the card’s conclusion limited to the visible conversations you inspected.

Save a comparison card and reopen the file

On Windows, open Start, search for Notepad and create a blank document. Copy the template below. Start with not checked in the observation fields, then replace only the statuses you actually inspected. Use short non-sensitive labels instead of full conversation text. If the operation stops, describe the particular missing input or control in the unresolved field.

Gemini branch comparison 02
Date and account type:
Original conversation title:
Recognition cue for selected earlier response:
Cue for the original later continuation:
Branch title and alternative-request cue:
Starting response observation: not checked
Original conversation reopened: not checked
Branched conversation reopened: not checked
Unresolved issue and next step:
Limit: not history restoration or proof of memory isolation.

Press Ctrl+S, choose Documents and save as gemini-branch-comparison-02.txt. Close the editor and reopen that file from the folder. Check that the starting-response cue, original later-continuation cue and alternative-request cue remain separately readable. Reopening the local card is a third check, distinct from reopening either conversation. It cannot stand in for either account observation.

A fifteen-minute planning allowance can cover identifying the boundary, requesting one alternative, checking both conversations and saving the card. This is not a measured completion-time promise. An unresolved card can still accurately identify the obstacle. Expanding the task into account maintenance or repeated unrelated prompts would make that obstacle harder to interpret, not turn it into a successful branch.

For other bounded outputs, the Canvas checklist walkthrough builds a prototype, while the scheduled-prompt guide manages a recurring request. Neither is evidence that this comparison passed. Close this exercise with the observed status of the original, the branch and the reopened card, without inventing an account result or claiming permanent preservation.